Author response: Mechanisms governing target search and binding dynamics of hypoxia-inducible factors
Year: 2022
Type: peer-review
Abstract: Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Appendix 1 Appendix 2 Data availability References Decision letter Author response Article and author information Metrics Abstract Transcription factors (TFs) are classically attributed a modular construction, containing well-structured sequence-specific DNA-binding domains (DBDs) paired with disordered activation domains (ADs) responsible for protein-protein interactions targeting co-factors or the core transcription initiation machinery. However, this simple division of labor model struggles to explain why TFs with identical DNA-binding sequence specificity determined in vitro exhibit distinct binding profiles in vivo. The family of hypoxia-inducible factors (HIFs) offer a stark example: aberrantly expressed in several cancer types, HIF-1α and HIF-2α subunit isoforms recognize the same DNA motif in vitro – the hypoxia response element (HRE) – but only share a subset of their target genes in vivo, while eliciting contrasting effects on cancer development and progression under certain circumstances. To probe the mechanisms mediating isoform-specific gene regulation, we used live-cell single particle tracking (SPT) to investigate HIF nuclear dynamics and how they change upon genetic perturbation or drug treatment. We found that HIF-α subunits and their dimerization partner HIF-1β exhibit distinct diffusion and binding characteristics that are exquisitely sensitive to concentration and subunit stoichiometry. Using domain-swap variants, mutations, and a HIF-2α specific inhibitor, we found that although the DBD and dimerization domains are important, another main determinant of chromatin binding and diffusion behavior is the AD-containing intrinsically disordered region (IDR). Using Cut&Run and RNA-seq as orthogonal genomic approaches, we also confirmed IDR-dependent binding and activation of a specific subset of HIF target genes. These findings reveal a previously unappreciated role of IDRs in regulating the TF search and binding process that contribute to functional target site selectivity on chromatin. Editor's evaluation This work on dissecting the function of transcription factors using single-molecule methods is likely to appeal to the broad eLife readership and the gene regulation community at large. In particular, the role of transcription activation domains in transcriptional specificity is a timely contribution to the field. https://doi.org/10.7554/eLife.75064.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Sequence-specific transcription factors (TFs) are key frontline regulators of gene expression. Classical LexA-Gal4 domain-swap experiments in yeast presented a simple modular structure and apparent division of labor for typical TFs (Brent and Ptashne, 1985). In this textbook paradigm, the DNA-binding domain (DBD) is responsible for DNA sequence recognition and binding specificity while the activation domain (AD) is responsible for target gene transactivation that involves protein-protein interactions with co-factors, the basal transcription machinery, and other ancillary factors that are generally devoid of sequence-specific DNA recognition. In higher eukaryotes, each DBD class usually contains multiple closely related family members. For example, the bHLH class of TFs includes MyoD, Clock, and Max. They all recognize the same E-box DNA-binding sequence motif 5'-CACGTG-3', yet each differentially regulates muscle differentiation, circadian rhythm, and cell proliferation, respectively (Kribelbauer et al., 2019). This raises the specificity paradox: how do TFs with seemingly identical DNA sequence specificity, at least as determined in vitro, nevertheless exhibit non-overlapping binding profiles in vivo and carry out distinct and even opposing functions? In general, when confronted with this conundrum, we have assumed that one or more co-factors or perhaps still to be identified 'silent partner' TFs can somehow divert target recognition to a composite cis-regulatory site distinct from the canonical DNA-binding site. Given the high occurrence of short binding motifs for most TFs throughout the genome, even with co-operative binding to composite sites, most potential specific binding sites nevertheless remain unoccupied as determined by genome-wide TF binding studies. What feature or motif within TFs outside of the DBD and dimerization domain may be responsible for such differential site selection has remained unclear. Thus, the simple rule of modular units with well separated divisions of labor between DBD, dimerization, and transactivation may deserve a closer look. We also wondered whether quantitative single molecule dynamics measurements might reveal new aspects of TF behavior in living cells that could inform us regarding potential mechanisms influencing the target search and binding process and differential site selectivity in a native physiologically relevant context. Here, we have chosen the hypoxia-inducible factors (HIFs) as a representative example to study the paradox of highly conserved DBDs carrying out distinct target site selection and to dissect potential novel features of TFs that mediate chromatin binding. HIFs are a family of α/β heterodimeric TFs stabilized under hypoxic conditions to promote angiogenesis, anaerobic metabolism, cell proliferation, and 'stemness' (Semenza, 2012). The oxygen-labile alpha subunits (mainly HIF-1α and HIF-2α) complex with their oxygen-stable beta partner (mainly HIF-1β) to form a functional dimer (Figure 1A). All HIF subunit isoforms belong to the bHLH-PAS (basic helix-loop-helix-PER-ARNT-SIM) family, where the N-termini are structured domains containing bHLH (DNA-binding) and PAS (dimerization) domains, while the C-termini consist of intrinsically disordered regions (IDRs) containing ADs (Figure 1A; Figure 1—figure supplement 1A). HIF-1α/1β and HIF-2α/1β dimers share a conserved structural fold (Wu et al., 2015), recognize the same hypoxia response element (HRE) 5'-TACGTG-3' binding motif (Schödel et al., 2011; Wenger et al., 2005), but share only a partial overlap of target genes in vivo (Smythies et al., 2019). With their own unique target gene sets, HIF-1α and -2α can exert divergent and even contrasting functions (Keith et al., 2012). For example, while both HIF-1α and HIF-2α regulate angiogenesis, HIF-1α specifically regulates glycolysis, apoptosis, and promotes NO production, whereas HIF-2α binds to the POU5F1 locus to maintain Oct4-regulated stem cell identity and pluripotency, promotes cell cycle progression, and inhibits NO production (Keith et al., 2012; Smythies et al., 2019). Therefore, our current simple textbook model of exchangeable modular TF functional units does not satisfactorily explain such isoform-specific target gene regulation. Figure 1 with 3 supplements see all Download asset Open asset Endogenous tagging of hypoxia-inducible factors (HIFs) in 786-O clear cell renal cell carcinoma (ccRCC) cells for fast single particle tracking (fSPT). (A) Schematic showing the similar domain organizations of HIFs (top) and the HRE-bound HIF α/β dimers (bottom). Disordered regions are represented as wavy lines. (B) Generation of Halo KIN clones in the HIF-1α negative 786-O ccRCC line. Top: Halo-tagging scheme of HIF-2α (left) and HIF-1β (right). Bottom: Western blot of wild-type (WT) 786-O cells and homozygously tagged knock-in clones (A31 and A21). Panels derive from the same blots shown in full in Figure 1—figure supplement 2. See Figure 1—source data 1 for images of the raw blots. (C) Halo-tagged HIF-2α and HIF-1β show predominant nuclear localization. Top: Schematic of labeling Halo-tagged proteins in live cells with cell-permeable Halo-binding JFX646 dye. Bottom: Representative images of Halo-HIF-2α (left) and Halo-HIF-1β (right) clones labeled with 500 nM JFX646 (D) representative images showing the same cell labeled with a high concentration of JFX549 dye for localizing the nucleus in one channel (left) and labeled sparsely with JFX646 dye for tracking individual molecules in another channel (right). Scale bar = 2 μm. (E) Graphical illustration of fSPT capturing trajectories of moving particles. Top: Highly inclined and laminated optical sheet illumination (HiLo). Bottom: Illumination and camera sequence with corresponding particle position at each frame (solid magenta dots). Particle's past positions (dashed magenta circles) are connected with dotted magenta lines to show the particle's trajectory. (F) Actual data showing detection of Halo-HIF-2α protein molecules at 5.48 ms frame rate. Scale bar = 2 μm. Figure 1—source data 1 Source data for both Figure 1 and Figure 1—figure supplement 2, including six raw images (tif files) for western blots and one figure (pdf) explaining each of the raw images. https://cdn.elifesciences.org/articles/75064/elife-75064-fig1-data1-v2.zip Download elife-75064-fig1-data1-v2.zip The HIF family differential specificity paradox is even more daunting to comprehend at the level of disease inducing mechanisms. HIFs are aberrantly upregulated and recognized as oncogenic drivers in multiple cancers. However, in addition to their shared roles in cancer onset and progression, HIF-1α and -2α also show many independent, sometimes even opposing roles in specific contexts (Keith et al., 2012). For example, in clear cell renal cell carcinoma (ccRCC), HIF-2α is the critical tumorigenic driver whereas HIF-1α, in contrast with its usual tumorigenic role, is mostly tumor-suppressive (Raval et al., 2005; Schödel et al., 2016). The regulatory mechanism behind such highly divergent outcomes is still largely unknown. Given such complexity, without a deeper understanding of isoform-specific transcriptional regulation, it is hard to predict the functional outcomes mediated by individual HIF isoforms in various cancer types or stages, which could be a complicating factor in developing more effective HIF-targeting cancer therapeutics. In this study, we aim to understand the molecular mechanisms mediating isoform-specific target gene regulation at its most fundamental level – could we detect differential molecular dynamics of distinct TF isoforms during the target search and chromatin binding process in live cells? Which regions or domains of TFs might be responsible for such isoform-specific properties? Could we begin to discern possible mechanisms that guide TFs with highly conserved DBDs to their distinct and specific targets beyond cognate DNA sequence recognition? Here, we use HIFs as an illustrative example, combining endogenous tagging and super-resolution single particle tracking (SPT) (Liu et al., 2015) to study the dynamic behavior of these key gene regulators in live cells under physiological conditions. We also dissect the contribution of different domains of HIF-α isoforms by a series of mutation and domain-swap experiments to directly test the concept of modular functional domains. Deploying a combination of genetic and small-molecule perturbations, we found that, although HIF DBD and dimerization are important for DNA target acquisition, differences between HIF isoforms in terms of amount of protein bound and diffusion characteristics are mainly driven by regions outside the DBD and dimerization domains. Finally, using genomic approaches we found that, in concordance with our imaging results, binding strength and gene activation are IDR-dependent for a subset of HIF target sites. Our results reveal a previously unappreciated role of unstructured domains in the target search and binding properties of TFs to functional chromatin sites in a live cancer cell context. Results Establishing a human cancer cell system for live-cell single molecule imaging of HIF To investigate HIF dynamics, we first focused on one of the cognate dimers: HIF-2α/1β. We used the common ccRCC line 786-O (Brodaczewska et al., 2016), derived from a VHL-deficient, H2 type primary ccRCC, wherein HIF-2α is stabilized due to an inactivating mutation in VHL (the E3 ubiquitin ligase that targets all HIF-α isoforms for proteasomal degradation) (Gnarra et al., 1994). The 786-O line also conveniently lacks any functional HIF-1α due to a truncating mutation of HIF-1α (Shen et al., 2011; Swiatek et al., 2020), which allows us to study one α isoform independently from the other. Using CRISPR/Cas9-mediated genome editing, we successfully generated several clonal lines with homozygous knock-in (KIN) of the HaloTag (Los et al., 2008) at the N-terminus of either HIF-2α or its binding partner, HIF-1β (Figure 1B; Figure 1—figure supplement 2). Western blotting confirmed that the tagged proteins are expressed at levels similar to wild-type (WT) in unedited cells (e.g. HIF-2α clone A31 and HIF-1β clone A21) (Figure 1B). Confocal imaging after covalently labeling cells with a fluorescent Halo-binding ligand (JFX646) (Grimm et al., 2021) shows the expected nuclear localization for both Halo-HIF-2α and Halo-HIF-1β proteins (Figure 1C). In addition, we confirmed by ChIP-seq that both tagged proteins maintain a similar genome-wide binding profile as the WT protein in unedited cells (Figure 1—figure supplement 3A), and RNA-seq confirmed that gene expression profiles in both edited cell lines are not significantly altered from WT unedited cells (Figure 1—figure supplement 3B–C, with cells overexpressing WT and mutant HIF-α as controls). We have thus established a human cancer cell system suitable for live-cell imaging of HIF-2α and HIF-1β at endogenous expression levels. To evaluate how HIF-2α and -1β explore the nucleus and bind DNA, we used the fast modality of super-resolution live-cell SPT (fast SPT, or fSPT) that is capable of tracking rapidly diffusing molecules. Cells with either HIF-2α or HIF-1β Halo KIN were doubly labeled with the live-cell permeable Halo-binding JFX dyes (Figure 1D) and were imaged under highly inclined and laminated optical sheet illumination (HiLo) (Tokunaga et al., 2008) at high frame rates (~182 Hz) to capture the movement of single molecules in their native nuclear environment (Figure 1E). Stroboscopic illumination at high excitation power is used to minimize motion blur (Elf et al., 2007; Hansen et al., 2017; Hansen et al., 2018), while sparse labeling ensures that only a limited number of molecules are detected at any given time in the nucleus to minimize misconnections when computing the path of individual molecules (trajectories) (Figure 1E–F). We can then estimate relevant kinetic parameters from these trajectories, extracting quantitative information such as diffusion coefficients and bound fraction. fSPT detects various HIF molecular states in their native nuclear environment To quantitatively analyze the acquired fSPT data, we used a non-parametric Bayesian 'state array' (SA) approach recently developed in our lab, which estimates the occupations of a range of dynamic states with diffusion rates from 0.01 to 100 μm2/s while accounting for known experimental biases due to localization error and fluorophore defocalization (Heckert et al., 2022). Briefly, SA analyzes regular Brownian motion with normally distributed localization error (RBME) with a two-dimensional 'grid of states' that spans a range of diffusion coefficients (first dimension) and localization error (second dimension) magnitudes, and evaluates the occupancy (proportion) of each state in this array using the observed fSPT trajectory data and a variational Bayesian algorithm. After inference, the proportion of molecules as a function of diffusion coefficient is calculated by marginalizing localization error, creating a 'diffusion spectrum' that reports the underlying molecular subpopulations according to their proportions (y-axis) and their diffusive properties (diffusion coefficients, x-axis) (Figure 2—figure supplement 1). While our previous approach, which already addressed biases due to localization error and defocalization, required the assumption of a fixed and limited (3 or less) number of states for the tracked protein (Hansen et al., 2018), our current SA approach does not require a priori knowledge or any assumption regarding the number of subpopulations or 'states' of the tracked molecules. We chose to use this new method, because as dimeric TFs, HIF subunits could conceivably exist in many states (e.g. bound, moving as a monomer, moving as a dimer, and moving in a bigger complex containing co-regulators) (Figure 2A–B) and the SA approach better suits our needs for model-free analysis without assuming any given number of states. Figure 2 with 3 supplements see all Download asset Open asset Fast single particle tracking (fSPT) sensitively detects molecules in a range of states. (A–B) Likelihood of diffusion coefficients based on a model of regular Brownian motion with normally distributed localization error (RBME) (Heckert et al., 2022), for (A) Halo-HIF-2α, clone A31 and (B) Halo-HIF-1β, clone A21, with drawing illustrating bound and different hypothetical moving states: complexes, dimer and monomer. Each row represents data collected from one cell. 0.1 μm2/s is used as the cut-off for bound versus free. (C) Top: Proportion of molecules as a function of their diffusion coefficients (posterior mean occupations for a state array; Heckert et al., 2022) evaluated on trajectories across all cells measured for each knock-in(KIN) line (Halo-HIF-2α, clone A31 and Halo-HIF-1β, clone A21). Compared to HIF-2α, HIF-1β has less bound fraction (gray versus yellow shaded areas) and faster diffusion coefficient (gray versus yellow arrows). Bottom: Summary of the bound fraction (left) and peak diffusion coefficient (right) for the two clones. Each bar represents the averaged value from three independent measurements on different days (black dots). Supplementary file 1 reports cell and trajectory numbers for this and all the following figures. (D–E) Overexpressing HIF-2α, but not a dimerization mutant form, in the Halo-HIF-1β knock-in (KIN) line increases HIF-1β binding and decreases its diffusion coefficient. (D) schematic illustrating the parental Halo-HIF-1β KIN cells (gray background) and cells stably overexpressing (O.E.) either the wild-type (darker blue background) or a dimerization mutant (DM, black crosses, lighter blue background) form of HIF-2α. (E) Top: Proportion of molecules as a function of diffusion coefficient measured for HIF-1β in Halo-HIF-1β KIN cells (gray) and in Halo-HIF-1β KIN cells overexpressing HIF-2α (WT, dark blue background, or dimerization mutant (HIF-2αDM), light blue). Shaded areas indicate bound fraction. Bottom: Bar plot of the average value (bar height) of the bound fraction (left) and peak diffusion coefficient (right) calculated from three independent measurements (black dots) for each condition. We first analyzed individual cells to check cell-to-cell variation. Due to the small number of trajectories per cell, we used a less precise version of the SA calculation, the 'naïve occupation estimate' by simply applying the RBME likelihood estimation to individual cells without refining the posterior diffusion coefficient and localization error (Figure 2—figure supplement (Heckert et al., 2022). We observed results for both HIF-2α and with diffusion coefficient estimates from cell to cell (Figure Figure 2—figure supplement clones A31 and A21). While the we observed likely due to the limited number of trajectories collected from each cell small to a in behavior of HIF proteins in each cell, our findings do indicate that a range of moving states likely for both HIF We then trajectories from many cells from three for each experimental to estimate the of diffusion coefficients for the SA a of diffusion coefficient estimates that reports the fraction of stably bound molecules while the full of the diffusing molecules (Figure Figure 2—figure supplement We the fraction with a diffusion coefficient μm2/s as and Figure 2—figure supplement 3 and Appendix 1 for of of between we observed a different behavior for Halo-HIF-2α to a fraction of Halo-HIF-2α is bound, the of Halo-HIF-1β diffusing (Figure Figure 2—figure supplement clones A31 and A21). the diffusion coefficient for the Halo-HIF-1β is that of We measurements in different KIN clones and confirmed the of these results for both Halo-HIF-2α (Figure 2—figure supplement clone and Halo-HIF-1β (Figure 2—figure supplement clone that although we quantitatively the bound due to the of the of the moving multiple states might exist but are a single quantitatively or mean diffusion coefficient may different Therefore, for the of the we only the peak diffusion coefficient in the but both peak diffusion coefficient and mean diffusion coefficient for the moving in Supplementary file The differences between HIF-2α and -1β at because one HIF-2α and HIF-1β to they exist as a However, the endogenous HIF-1β is expressed at a higher level HIF-2α (Figure 1—figure supplement the of HIF-1β is likely to without HIF-2α. the plot only the fraction of molecules as a function of their diffusion but does not on the number of molecules. Therefore, a bound fraction for Halo-HIF-1β does not mean numbers of bound molecules Halo-HIF-2α, many more Halo-HIF-1β molecules are in the Given this we that HIF-1β molecular dynamics and binding be by the stoichiometry. HIF-1β binding and diffusing dynamics can be by and are on dimerization To test the that HIF-1β dynamics on we first to its behavior by stably overexpressing HIF-2α in the HIF-1β Halo-tagged KIN line (Figure We found that the of Halo-HIF-1β more when HIF-2α is most likely due to its dimerization with the HIF-2α to form dimers capable of binding. we also observed a in the Halo-HIF-1β bound fraction to (Figure Figure 2—figure supplement and To that the in HIF-1β dynamics by levels of HIF-2α are on we stably the HIF-2α mutant that previously to its dimerization with HIF-1β (Wu et al., of HIF-2α not the bound fraction or the diffusion coefficient of Halo-HIF-1β to the same with WT HIF-2α (Figure Figure 2—figure supplement that the we are We our results by stably overexpressing different of HIF-α in the HIF-1β Halo-tagged KIN line and cells with an small-molecule inhibitor, inhibits but not dimerization by specifically binding to the dimerization domain of HIF-2α (Figure supplement and thus has used as an HIF-2α for ccRCC et al., et al., 2019). We first confirmed that inhibits HIF-2α transcription function in a (Figure supplement 1B). also the HIF-2α bound fraction in the HIF-2α Halo-tagged KIN line in a similar the potential of fSPT to TF dynamics and functional (Figure supplement We to use for all experiments to its we out a series of experiments to probe the of different functional domains of HIF-1α and HIF-2α to which of these closely related TFs might be in when paired with Using the HIF-1β Halo-tagged KIN line as the parental we stably WT or where we the structured and disordered domains between HIF-1α and HIF-2α (Figure All these different HIF-α are expressed from a and are that only is used as We then these cells with either or and measured Halo-HIF-1β dynamics (Figure While overexpressing of which form, is to both the bound fraction and the diffusion coefficient of HIF-1β (Figure and Figure For cells overexpressing the α that the HIF-2α structured this on HIF-1β can be at least after (Figure 3B–C, and In for cells overexpressing the α that the HIF-1α structured this is to with the subunit isoform specificity of the drug for HIF-2α (Figure 3B–C, and In and only overexpressing with only the HIF-1β bound that the of HIF-1β is not with (Figure 3B–C, parental cell and these results that HIF-1β dynamics change after with its α partner and can be with a specific dimerization The observed differences also that fSPT is a to molecular dynamic of TFs in living cells us to new while we various perturbations, such as subunit concentration or and specific Figure 3 with 2 supplements see all Download asset Open asset HIF-α increases HIF-1β binding and decreases HIF-1β diffusion coefficient dimerization, in an intrinsically disordered region (A) Schematic of Halo-HIF-1β behavior with genetic and small-molecule Halo-HIF-1β knock-in (KIN) cells (gray background) and cells stably overexpressing (O.E.) either a certain form of HIF-α or domain HIF-1α, HIF-2α, Disordered regions are represented as wavy or a only background) are with and without dimerization treatment. (B) Proportion of Halo-HIF-1β molecules as a function of diffusion coefficient measured in various conditions in (A) showing overexpressing α subunit can change HIF-1β Cells overexpressing the α subunit containing HIF-2α disordered region have a HIF-2α and to containing HIF-1α disordered region HIF-1α and and of HIF-1β as a function of diffusion measured in each of the six cell lines with are to the by overexpressing an α subunit can be specifically by for cell lines an α subunit that contains the HIF-2α structured domain (C) Summary of the average bound (top) and peak diffusion coefficient for all with black from each of the three individual outside the domain HIF molecular dynamics the effects of the different α variants, we found that of their structured with the same IDRs (Figure 3B–C, and the containing the HIF-2α have a on HIF-1β binding the containing the HIF-1α Thus, it that the bound fraction of HIF-1β is not determined by the HIF-α DBD, but by HIF-α which we found To the of HIF-α IDRs in HIF we a version of either HIF-1α or -2α that contains only the structured region or HIF-2α which still both the DBD and dimerization for with HIF-1β (Wu et al., both these the of HIF-α the HIF-1β bound fraction (Figure supplement 2). these HIF-α also only the HIF-1β diffusion coefficient. These results indicate that
Source: eLife
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,500
OpenAlex ID: W4312574357